Module learn 

Learn Module - 学習系 Domain の分離

設計思想

Engine (Core) と Learn の責務を明確に分離する。

• Core (Engine): ActionEvent を発行、基本統計を管理
• Learn: ActionEvent を受け取り、学習パターンを分析

Model Layer

Model (base)
  ├── Scorable      → 行動選択に使用（UCB1, Thompson, Greedy）
  └── Parametric    → 戦略設定に使用（Orchestrator初期化）

モデル種類

モデル	責務	生存期間
LearnedModel	行動選択スコア（遷移、N-gram等）	1セッション
OptimalParamsModel	パラメータ最適化（ucb1_c等）	複数セッション

LoRA

LoRA学習・適用は lora モジュールで管理。

• LoraTrainer: LoRA学習の実行
• TrainedModel: 学習済みLoRAアダプタ
• ModelApplicator: llama-server への適用

Re-exports

pub use stats_model::param_keys;

pub use stats_model::Model;

pub use stats_model::ModelMetadata;

pub use stats_model::ModelType;

pub use stats_model::ModelVersion;

pub use stats_model::OptimalParamsModel;

pub use stats_model::ParamValue;

pub use stats_model::Parametric;

pub use stats_model::Scorable;

pub use stats_model::ScoreContext;

pub use stats_model::ScoreModel;

pub use stats_model::ScoreQuery;

pub use stats_model::StatsModelId;

pub use snapshot::merge_snapshots;

pub use snapshot::FileSystemStorage;

pub use snapshot::LearningSnapshot;

pub use snapshot::LearningStore;

pub use snapshot::MergeStrategy;

pub use snapshot::SessionId;

pub use snapshot::SnapshotKey;

pub use snapshot::SnapshotMetadata;

pub use snapshot::SnapshotStorage;

pub use snapshot::TimeSeriesQuery;

pub use snapshot::Timestamp;

pub use snapshot::SNAPSHOT_VERSION;

pub use offline::ActionOrderSource;

pub use offline::LearnedActionOrder;

pub use offline::OfflineAnalyzer;

pub use offline::OfflineModel;

pub use offline::OptimalParameters;

pub use offline::RecommendedPath;

pub use offline::StrategyConfig;

pub use record::ActionRecord;

pub use record::DependencyGraphRecord;

pub use record::FromRecord;

pub use record::LlmCallRecord;

pub use record::Record;

pub use record::RecordStream;

pub use episode::Episode;

pub use episode::EpisodeBuilder;

pub use episode::EpisodeContext;

pub use episode::EpisodeId;

pub use episode::EpisodeMetadata;

pub use episode::EpisodeTrait;

pub use episode::Outcome;

pub use training::ConversationData;

pub use training::ConversationRole;

pub use training::ConversationTurn;

pub use training::TrainingData;

pub use training::TrainingFormat;

pub use training::TrainingMetadata;

pub use store::DefaultEpisodeRepository;

pub use store::EpisodeDto;

pub use store::EpisodeFilter;

pub use store::EpisodeMeta;

pub use store::EpisodeRepository;

pub use store::EpisodeStore;

pub use store::FileEpisodeStore;

pub use store::FileLoraStore;

pub use store::FileRecordStore;

pub use store::InMemoryEpisodeStore;

pub use store::InMemoryLoraStore;

pub use store::InMemoryRecordStore;

pub use store::LoraModelStore;

pub use store::ModelFilter;

pub use store::ModelMeta;

pub use store::OutcomeFilter;

pub use store::RecordFilter;

pub use store::RecordId;

pub use store::RecordMeta;

pub use store::RecordStore;

pub use store::RecordStoreError;

pub use store::StoreError;

pub use store::TrainedModelDto;

pub use store::TrainingMetricsDto;

pub use learn_model::DependencyGraphLearnModel;

pub use learn_model::DpoConfig;

pub use learn_model::DpoLearnModel;

pub use learn_model::DpoPair;

pub use learn_model::LearnError;

pub use learn_model::LearnModel;

pub use learn_model::WorkerDecisionSequenceLearn;

pub use learn_model::WorkerTaskLearn;

pub use trigger::AlwaysTrigger;

pub use trigger::AndTrigger;

pub use trigger::CountTrigger;

pub use trigger::ManualTrigger;

pub use trigger::NeverTrigger;

pub use trigger::OrTrigger;

pub use trigger::QualityTrigger;

pub use trigger::TimeTrigger;

pub use trigger::TrainTrigger;

pub use trigger::TriggerBuilder;

pub use trigger::TriggerContext;

pub use trigger::TriggerError;

pub use trigger::TriggerMetrics;

pub use lora::ApplicatorError;

pub use lora::LlamaServerApplicator;

pub use lora::LlamaServerConfig;

pub use lora::LoraModelId;

pub use lora::LoraTrainer;

pub use lora::LoraTrainerConfig;

pub use lora::LoraTrainerError;

pub use lora::ModelApplicator;

pub use lora::NoOpApplicator;

pub use lora::TrainedModel;

pub use lora::TrainingMetrics;

pub use daemon::Applier;

pub use daemon::ApplierConfig;

pub use daemon::ApplierError;

pub use daemon::ApplyMode;

pub use daemon::ApplyResult;

pub use daemon::DaemonBuilder;

pub use daemon::DaemonConfig;

pub use daemon::DaemonError;

pub use daemon::DaemonStats;

pub use daemon::DataSink;

pub use daemon::DataSinkError;

pub use daemon::DataSinkStats;

pub use daemon::LearningDaemon;

pub use daemon::ProcessResult;

pub use daemon::Processor;

pub use daemon::ProcessorConfig;

pub use daemon::ProcessorError;

pub use daemon::ProcessorMode;

Modules

daemon

Learning Daemon - 継続的学習プロセス

episode

Episode - 学習の基本単位

learn_model

LearnModel - 学習の統合モデル

lora

LoRA Module - LoRA学習・適用

offline

Offline Learning - セッション間学習の分析・最適化

record

Record - 生イベントの抽象化

snapshot

Learning Snapshot - 学習データの永続化システム

stats_model

Stats Model Layer - 行動選択用統計モデル

store

Store Module - 永続化レイヤーの Clean Architecture 実装

training

TrainingData - LoRA 学習用データ形式

trigger

TrainTrigger - 学習開始条件の判定

Structs

ConfidenceMapProvider

HashMap<String, f64> ベースの静的 Provider

ContextualActionStats

コンテキスト条件付きアクション統計

EpisodeTransitions

エピソード遷移統計

LearnStats

学習用統計の集約

LearnStatsProvider

LearnStats と ScoreModel を一体管理する Provider

NgramStats

N-gram 統計（3-gram, 4-gram パターン学習）

NullProvider

ボーナスを返さない Null Provider

ScoreModelProvider

ScoreModel を使った Provider 実装

SelectionPerformance

Selection 戦略効果測定（Meta-learning）

StrategyStats

戦略ごとの統計

StrategySwitchEvent

戦略切り替えイベント

Enums

LearningQuery

学習データへのクエリ

LearningResult

クエリ結果

Traits

LearnedProvider

学習済みデータへのアクセス Provider

Type Aliases

SharedLearnedProvider

Provider の共有参照型